Present address: State University of New York at Buffalo, Buffalo, NY 14260, USA.
Mei Guo and Mary A. Rupe have contributed equally to this work.
We have taken a genomic approach to examine global gene expression in the maize endosperm in relation to dosage and parental effects. Endosperm of eight hybrids generated by reciprocal crosses and their seven inbred parents were sampled at three developmental stages: 10, 14, and 21 days after pollination (DAP). These samples were subjected to GeneCalling, an open-ended mRNA-profiling technology, which simultaneously analyzes thousands of genes. Results indicated that the overall level of gene expression in the maize endosperm was dosage-dependent, that is, the gene expression was proportional to the parental genome contribution of 2n maternal : 1n paternal. However, approximately 8% of the genes deviated from such allelic additive expression and exhibited differential expression in hybrids of reciprocal crosses, resembling either maternally or paternally expressed genes. There were more genes with maternal-like expression (MLE) than those with paternal-like expression (PLE). Allele-specific expression analysis of four selected genes using the WAVE denaturing HPLC (dHPLC) system revealed several mechanisms responsible for the deviation from the allelic additive expression in the hybrid endosperm: heterochronic allelic variation, allelic variation in the level of expression, and genomic imprinting. We discovered a novel imprinted gene no-apical-meristem (NAM) related protein1 (nrp1) that was expressed only in the endosperm and regulated by gene-specific imprinting. The nrp1 gene, a putative transcriptional factor, may play an important role in endosperm development.
The double fertilization process of flowering plants results in the diploid embryo and the triploid endosperm, which are otherwise genetically identical. The endosperm has two maternal genomes (2n) and one paternal genome (1n). Such a unique genetic composition is important for normal seed development (Lin, 1984; Rhoades and Dempsey, 1966; Scott et al., 1998).
Limited information is available regarding the differences between the parental genomes at the gene-expression level during endosperm development. Many genes expressed in the endosperm are under the control of parent-of-origin effect. Parent-of-origin effect refers to different mechanisms that regulate gene expression as a result of the parental source. These mechanisms include maternal effects resulting from mitochondrial, chloroplast genomes, and other cytoplasmic factors, gene dosage effect and genomic imprinting, in which expression of the allele is dependent on maternal or paternal transmission. Although the 2n maternal : 1n paternal genome ratio is important for endosperm development, whether the parental genomes are expressed according to the same ratio of 2 : 1 at the transcript level (allelic dosage expression) is not known.
In addition to the effect on the level of gene expression, the parent-of-origin effect can also be manifested at the time of gene expression. Vielle-Calzada et al. (2000) demonstrated that during early seed development, the paternal genome exhibited delayed activation compared to the maternal genome. However, whether the parental genomes are differentially regulated throughout endosperm development remains to be revealed.
In this study, we examined gene expression during maize endosperm development in relation to the dosage and origin of the parental genomes using GeneCalling technology (Shimkets et al., 1999). The GeneCalling technology is an open-ended, gel-based method that reproducibly quantifies the mRNA level for thousands of genes simultaneously. This technology detects 80–90% of expressed genes in a given tissue. GeneCalling has been successfully used in identifying genes involved in the flavonoid pathways, root-lodging resistance (Bruce et al., 2000, 2001), and stress response during seed maturation and germination (Kollipara et al., 2002) in maize. We used this technology to analyze the mRNA expression in the maize endosperm of reciprocal hybrids and inbred parents. The majority of the genes in the endosperm were expressed in a dosage-dependent manner; however, deviations from such an allelic dosage expression were also observed. Selected genes representing allelic and non-allelic additive expression were further analyzed at an allele-specific level by using the WAVE denaturing HPLC (dHPLC) system. Such analysis led us to discover allelic expression variation resulting from different regulatory mechanisms in the maize endosperm and a new endosperm-specific gene, no-apical-meristem (NAM) related protein1 (nrp1), a putative transcription factor that showed gene-specific imprinting.
GeneCalling technology is based on restriction digestion and PCR amplification of cDNA fragments (Shimkets et al., 1999). cDNA is digested with restriction enzymes and amplified with 48 pairs of PCR primers, which covers 80–90% of the expressed genes represented in the cDNA pool from the tissue analyzed. One of the PCR primers is labeled with fluorescamine (FAM). The PCR fragments are separated on acrylamide gels and analyzed by electrophoretic scans. The fluorescent intensity from FAM-labeled cDNA fragments is proportional to the abundance of the corresponding mRNA expressed in the given tissue sample. Poly(A)+ RNA from endosperm tissues of eight hybrids and seven inbred parents were subjected to GeneCalling analysis as described by Shimkets et al. (1999; Experimental procedures). mRNA profiles of hybrids and inbreds were obtained for 10, 14, and 21 days after pollination (DAP) endosperm. PCR amplification using 48 primer pairs produces approximately 22 000 cDNA fragments for a given mRNA sample. More than one cDNA fragment could correspond to one gene.
Allelic additive gene expression
In order to obtain a quantitative measurement of the F1 hybrid expression level relative to the average of the parental levels (allelic additive expression level) for a corresponding cDNA fragment, we adapted the terminology ‘d/a’ ratio from quantitative genetics as a metric, where ‘d’ stands for a dominant gene action and ‘a’ stands for an additive gene action (Comstock and Robinson, 1952; Gardner et al., 1953; Experimental procedures). An additive gene action would predict the expression level of the gene in the F1 hybrid to be equal to the average of the parents. In the triploid endosperm, if gene expression is allelic additive, the expected expression level of individual genes in the hybrid would be proportional to the parental contribution, i.e. the average of parents (Ave) = (2Pfemale + 1Pmale)/3. Therefore, d = F1 hybrid − Ave; a = Parent 1 (P1) − Ave. The d/a ratio provides a quantitative metric to measure the level of gene expression in the hybrid in relation to the allelic dosage. While the absolute value of the d/a ratio indicates the degree of the deviation from the allelic additive expression, the sign of the d/a ratio indicates the direction of the deviation, maternal or paternal.
We calculated the d/a ratio for each cDNA fragment and estimated the overall gene expression in the endosperm of eight hybrids (Table 1) in relation to the allelic dosage expression. If gene expression in the endosperm were mostly dosage-dependent, we would expect that the majority of the cDNAs have a d/a ratio near 0 (because F1 hybrid = Ave), and thus the proportion of the cDNA fragments based on the d/a ratios would exhibit a normal distribution with the peak at 0. As shown in Figure 1, this was indeed the case. The same expression pattern was observed across all three developmental stages analyzed. These results suggested that allelic additive expression was the norm throughout seed development.
Table 1. List of inbredsa used to produce hybrids of reciprocal crosses and their characteristics
The seven inbred lines were crossed reciprocally to produce eight hybrids and self-pollinated to reproduce the inbred lines.
Public US Iowa Stiff Stalk Synthetic (100%), Reid yellow dent (YD)-type
Public US (YD) Lancaster Sure Crop (50%), Krug (50%)
Public line from Iowa Stiff Stalk Synthetic (100%), Reid yellow dent (YD)-type
Mid-maturity Non-Stiff Stalk (YD) type, not related to Mo17, Central US adaption
Mid-maturity Stiff Stalk (YD) type, Central/Eastern US Corn Belt adaption
Mid-maturity Non-Stiff Stalk (YD) type, not related to Mo17, Central US adaption
Mid-maturity Non-Stiff Stalk (YD) type, not related to Mo17, Central US adaption
Deviation of gene expression from the allelic dosage in the hybrid endosperm
GeneCalling technology is based on restriction digestion of cDNA fragments. Therefore, cDNA fragment differences between genotypes could be attributed to two sources: (i) differential mRNA expression and (ii) allelic sequence polymorphism (Bruce et al., 2001; Shimkets et al., 1999). As reciprocal hybrids have the same genetic constitution, comparing mRNA profiles of both hybrids and corresponding parents can reveal gene-expression differences in hybrids in relation to the allelic dosage. This technology also allows differentiating expression of maternal and paternal alleles in the hybrid when allele sequence polymorphisms exist.
The global gene expression in the endosperm suggested that a majority of the genes in the endosperm were expressed in an allelic dosage manner (Figure 1). To determine what proportion of the genes violated the allelic dosage prediction, we took advantage of the technology and used allele sequence polymorphisms to measure allele-specific expression affected by maternal or paternal transmission. We compared mRNA profiles of the eight hybrids and corresponding parents using the selection criteria described in Experimental procedures and Figure 2(a). The selection was based on a minimum twofold change in the transcript level. Depending on the expression level of genes, a lower-fold change cut-off would favor the false-positive rate and a higher-fold change cut-off would favor the chance of missing differentially expressed genes. Based on CuraGen's GeneCalling analysis on the reproducibility, sensitivity, and false-positive rate of the technology (Shimkets et al., 1999), we chose twofold as an empirical cut-off. The twofold cut-off is a general protocol that has been used in other studies (Bruce et al., 2000, 2001). In this analysis, expression levels that varied in at least twofold were therefore considered as different between compared samples. Also, the expression level in this study was limited to reflect the steady state mRNA level only.
cDNA fragments were classified as maternal-like expression (MLE) or paternal-like expression (PLE), respectively, based upon the F1's expression pattern, resembling either the maternal or paternal parent. The expression patterns of these cDNA fragments were visualized in GeneScape, a GeneCalling interface for expression profile analysis (Shimkets et al., 1999). Examples of the GeneCalling traces of the cDNA fragments of MLE and PLE are shown in Figure 2(b,c).
The fraction of cDNA fragments that were expressed differentially between reciprocal hybrids and deviated from the allelic dosage expression (MLE or PLE) was 8.2% at 10 DAP, 9.0% at 14 DAP, and 7.0% at 21 DAP, based on the average of eight hybrids (Table 2; Figure 3). Within one developmental stage, the percentage of genes that exhibited such an expression pattern varied among different genotypes (e.g. at 10 DAP, the PLE was 7.1% in hybrid S1/N1 versus 1.2% in Mo17/B73). There was also variation through the developmental stages within an individual genotype, such as in hybrid S1/N1, the PLE was 7.1% at 10 DAP versus 1.9% at 21 DAP.
Table 2. The number of cDNA fragmentsa differentially expressed between parents and hybrids
Numbers in the table are average values of eight hybrids. The percentages were based on the total cDNA fragments that were different between the parents.
Allele-specific analysis of gene expression
To identify genes that exhibited different patterns of expression in GeneCalling, we requested cloning and sequencing of a number of cDNA fragments representing each category of allelic dosage expressed, MLE and PLE. We were successful in obtaining cDNA fragments of one allelic dosage expression, seven MLEs, and no PLE. Because we were limited with the number of cDNA fragments we could isolate from CuraGen and the access of the GeneCalling technology, we were able to pursue only these eight cDNA fragments. In addition, isolation of the cDNA fragments of low abundance is less successful than those of high abundance with this technology.
Eight sequenced cDNA fragments corresponded to five genes based on their sequence homology to expressed sequence tags (ESTs) in Pioneer/DuPont databases. The allelic-dosage-expressed cDNA was a novel gene and was used as a control in allele expression analysis. Among MLE genes, we found 18-kDa δ-zein (dz-18, two cDNA fragments) and 10-kDa δ-zein (dz-10, one cDNA fragment). The 18- and 10-kDa δ-zeins are storage methionine-rich proteins which are highly transcribed in the maize endosperm (Chui and Falco, 1995; Kirihara et al., 1988; Woo et al., 2001). The third MLE gene was identified as β-glucosidase aggregating factor (BGAF, one cDNA fragment). BGAF has previously been cloned and is responsible for the β-glucosidase null phenotype in maize by interacting with β-glucosidase and forming insoluble aggregates (Blanchard et al., 2001; Esen and Blanchard, 2000).
The most important discovery was a new endosperm-specific gene (three cDNA fragments), which we named nrp1. This gene encoded a protein that shared a high homology with the NAM protein from Petunia. The Petunia NAM mutant failed to develop a shoot apical meristem (Souer et al., 1996).
To confirm the MLE pattern of expression for each individual gene, we used gene-specific primers to obtain the cDNA from each inbred parent by RT-PCR. We then sequenced the PCR products to identify allele sequence polymorphisms that would allow separation of the two parental alleles on the WAVE dHPLC system (Transgenomic, Omaha, NE, USA). Table 3 summarizes the data used in the WAVE dHPLC analysis. Primers were designed in consensus regions between the parental alleles to avoid amplification preference to either allele and to optimize the amplicon for analysis on the WAVE dHPLC. For the novel gene, we could not determine if the sequences of the primer regions between the parental alleles were conserved. Instead, we tested these primers with genomic DNA extracted from the endosperm, and no amplification bias was observed. The PCR products from hybrid cDNA were then subjected to the WAVE dHPLC for separation.
Table 3. The five genes characterized using the WAVE dHPLC system
Allelic sequence polymorphism used in WAVE analysisb
The size of the cDNA fragment amplified by RT-PCR. The size of the longer allele fragment is shown when there was a length polymorphism.
b (-) indicates deletion/insertion between the B73 and Mo17 alleles. For dz-18, SNPs (underlined) across the amplified region were used. ‘x’ represents the non-polymorphic regions (not the actual number of the nucleotides) between the alleles.
When two types of cDNA sequences were present in one sample (e.g. both alleles were expressed in the hybrid), in addition to the homoduplexes, heteroduplexes were formed. Chromatogram traces for each PCR were generated by UV detection. Peak areas corresponding to the homoduplexes and heteroduplexes were calculated by the wavemaker software (Transgenomic, Omaha, NE, USA) and used for allele-dosage-expression analysis.
In order to discriminate family members, we designed member-specific primers when the member sequences were available. We further sequenced the PCR products of each primer set, and in each case, we only found one product, indicating the primers to be specific to the gene. We also performed blast searches with the primer sequences in Pioneer/DuPont databases and confirmed that the primer sets did not have any significant match that would amplify other genes.
These five genes were subjected to allele expression analysis with the WAVE dHPLC system. We measured the relative expression level of the parental alleles in the hybrids. Two sets of reciprocal hybrids of different genetic backgrounds were examined for each gene, except for dz-10, in which a resolvable allelic sequence polymorphism could not be found in parents of the second hybrid. The expression patterns of these tested genes in GeneCalling in one hybrid are shown in Figure 4. The allele-specific expression results from the WAVE dHPLC (Figure 5) were consistent with the observation in GeneCalling. Because 30-cycle PCR was used in the WAVE dHPLC as compared to 20-cycle PCR in GeneCalling, some allele expression under the detection limit by GeneCalling could be revealed by the WAVE dHPLC analysis.
Allele-specific analysis and quantification of the novel gene representing the allelic dosage expression is shown in Figure 5 and Table 4. The maternally contributed allele expressed twice as much as the paternal allele, and the relative level of the alleles fitted the ratio of 2 maternal : 1 paternal throughout the developmental stages. The expression in both reciprocal hybrids was consistent with the prediction of allelic dosage. The same allelic dosage expression pattern of this gene was found in a hybrid of different genetic backgrounds.
Table 4. Ratios of maternal to paternal allele expression in relation to allelic dosage in the endosperm
Varied allele expression
A novel gene
Mean values are ratios of the maternal to paternal allele (M/P). The M/P ratio was calculated using allele peak areas, and compared with the predicted ratio of 2 maternal : 1 paternal according to allelic dosage gene expression. The mean and the standard error (SE) based on three replicates of the WAVE dHPLC data are shown. Results with P < 0.01 in t-test are shown in bold and italic. NA (not applicable) were cases where there was no detectable expression of the paternal allele and the denominator of maternal:paternal was 0. (–) denotes that no tissue sample were analyzed. B: B73; M: Mo17.
Samples collected in year 1998. All other samples were collected in year 2001.
For nrp1 only, hybrid genotypes were S1S1/N1 and N1N1/S1.
The MLE cDNA fragments corresponding to BGAF, dz-18, and dz-10 exhibited expression patterns that were distinct from the allelic dosage expression of the novel gene (Figure 5; Table 4). First, these genes exhibited a heterochronic allelic variation. Upon maternal transmission, the Mo17 and N2 alleles exhibited a delay in expression as compared to the B73 and S2 alleles. The level of Mo17 and N2 allele expression was near background at 10 DAP in some cases, whereas the B73 and S2 alleles expressed predominantly throughout the developmental stages. At later stages, the expression of Mo17 and N2 alleles progressively increased, and in dz-10, the Mo17 allele expressed at an equivalent level as the B73 allele, exhibiting the allelic dosage expression by 21 DAP. Similar patterns of heterochronic allele variation were observed among all three genes. The allele-expression patterns were generally the same between the two hybrids of different genetic backgrounds. The heterochronic allelic variation of the Mo17 and N2 alleles was more evident upon maternal transmission than paternal transmission.
Second, the Mo17 and N2 alleles differed from the B73 and S2 alleles in the level of expression in BGAF and dz-18 (Figure 5; Table 4). The Mo17 and N2 alleles appeared to be weaker-expressing alleles and were expressed at a lower level as compared to B73 and S2 alleles. The ratio of B73:Mo17 tended to be higher than 2 : 1 in B73/Mo17 and the ratio of Mo17:B73 tended to be lower than 2 : 1 in Mo17/B73. In some cases (e.g. dz-18), the maternally transmitted Mo17 and N2 alleles expressed at a level lower than the paternal B73 and S2 alleles, despite having two doses. Upon paternal transmission, the Mo17 and N2 alleles generally had no or very low expression. The Mo17 allele of dz-10 was not a weaker-expressing allele by nature as compared to the B73 allele because the expression level of the Mo17 allele was equivalent to that of the B73 allele, i.e. consistent with the allelic dosage (2Mo17 : 1B73) in hybrid Mo17/B73 at 21 DAP.
The pattern of expression of the nrp1 gene was distinct from the heterochronic allelic expression discussed above. The NRP1 protein belongs to a family of NAM transcription factors as shown by the alignment of the amino acid sequence with Petunia NAM and two rice homologs OsNAC1 and OsNAC2 (Figure 6a). The overall amino acid identities of NRP1 with OsNAC1, OsNAC2, and NAM were 34, 35, and 32%, respectively. A higher percentage of amino acid identity was found at the five N-terminal NAC domains: 55, 58, and 59% with NAM, OsNAC1, and OsNAC2, respectively. The NAC domains are highly conserved among the NAM/NAC family members (Kikuchi et al., 2000).
The expression pattern of the nrp1 gene indicated that nrp1 was an imprinted gene (Figure 6b; Table 4). The ratio of maternal allele:paternal allele expression level deviated significantly from 2 : 1. While the maternally transmitted alleles were expressed, the paternally transmitted alleles were nearly silenced throughout the developmental stages analyzed. The silencing of the paternal allele was more in the hybrid of B73 and Mo17 than in the hybrid of S1 and N1. However, nrp1 exhibited genomic imprinting in both hybrids of different genetic backgrounds. This result suggested that the imprinting was the nature of the gene and not the nature of the allele.
The nrp1 gene was expressed specifically in the endosperm. Thirty-cycle RT-PCR of RNA isolated from the root, stalk, mature and immature leaf, immature ear, tassel, endosperm, embryo, and ovule did not detect the expression of nrp1 in any of the tissues tested except for the endosperm (Figure 6c). The lack of expression in the ovule indicated that nrp1 was expressed only after fertilization, and expression of the maternal allele could not be explained by cytoplasmic effect or transcript present prior to fertilization.
The nrp1 expression during kernel development was analyzed using the massively parallel signature sequencing (MPSS; Brenner et al., 2000a,b). With the MPSS technique (Figure 6d), each cDNA is attached to the surface of a unique microbead. A highly expressed mRNA is represented on a proportionately large number of microbeads. Signature sequences of 17 nucleotides are then obtained from these microbeads by iterative cycles of restriction with a type IIs endonuclease, adaptor ligation, and hybridization with encoded probes. The technique provides an unprecedented depth and sensitivity of mRNA detection, including very low-expressed messages. The level of expression (p.p.m.) of a gene is determined by the abundance of its signature in the total pool. The nrp1 gene was expressed throughout endosperm development, peaking at 25 DAP, and was not detected in the ovule. These results also indicate that the nrp1 gene was expressed only after fertilization occurred.
Previously, very little was known about the global gene expression patterns in the plant endosperm. Our studies of genome-scale mRNA-profiling analysis of the hybrid and inbred endosperm indicated that the majority of genes in the triploid endosperm expressed in an allelic additive manner. We also found that approximately 8% of the profiled cDNAs were differentially expressed between hybrids of reciprocal crosses and did not fit allelic dosage expression. As the GeneCalling data were based on three PCR experiments for an RNA sample with no RNA sample replication, some of the variation could be attributed to noise, especially for genes expressed at a low level. Analysis of three RNA sample replicates from one genotype, however, showed little variation among replications. We verified the expression patterns of eight individual cDNA fragments by using the WAVE dHPLC system and obtained consistent results as compared to GeneCalling profiles. Allele expression analysis of the MLE cDNA fragments suggested that three mechanisms were responsible for the non-allelic dosage expression in the hybrid endosperm: (i) heterochronic allelic variation, (ii) allelic variation in the level of expression, and (iii) genomic imprinting.
Allelic dosage expression
The allelic dosage expression represents the expression pattern of the majority of endosperm genes. Our results are consistent with the notion that for most genes, both maternally and paternally transmitted alleles may be expressed at equivalent levels (Alleman and Doctor, 2000), and dosage effects play an important role in endosperm development (Baroux et al., 2002; Birchler, 1993). Our results also provide the molecular evidence to correlate with the requirement of a strict maternal/paternal ratio for normal endosperm development. Studies of maize and Arabidopsis endosperm development suggest that the endosperm develops normally only when the ratio of parental contributions is 2 maternal : 1 paternal. The further the deviation is from the 2 : 1 ratio, at least up to a certain ploidy level, the more defective is the endosperm (Alleman and Doctor, 2000; Lin, 1984; Rhoades and Dempsey, 1966; Scott et al., 1998). These results cannot be explained solely by gene imprinting, i.e. simple inactivation of either all maternal or all paternal alleles of genes in the process (Berger, 1999; Birchler, 1993; Birchler and Hart, 1987). Parental genome balance may be necessary, and at least some of the genes from both parental genomes would be required to function.
Heterochronic variation and allelic expression level variation
Nevertheless, there were genes that deviated from the allelic dosage expression in the maize endosperm. Allele-specific expression analysis of BGAF, dz-18, and dz-10 indicated allelic variation in timing and in the level of expression. Both types of allelic variations can be attributed to the deviation from allelic dosage expression of dz-18 and BGAF. However, the non-allelic dosage expression patterns of dz-10 appeared to be because of the heterochronic allelic variation only, and did not involve allele expression level variation. The delayed expression of the Mo17 and N2 alleles of these genes was also responsible for the low or no detected expression of these alleles at earlier stages.
The heterochronic allelic variation in timing of expression was recently reported by Cong et al. (2002) in the tomato fw2.2 gene, a putative negative regulator of cell division in fruit development. The heterochronic allelic variation and total transcript level of the fw2.2 gene are associated with tomato fruit size variation. These results provide molecular evidence that the allelic variation in expression level and timing can be accountable for the phenotypic differences. The significance of heterochronic allelic variation of the BGAF, dz-18, and dz-10 in the hybrid endosperm is unclear. The major role of maize BGAF appears to be in the defense and stress response (Blanchard et al., 2001; Esen and Blanchard, 2000), and the zeins are major seed-storage proteins. As the endosperm is the major food resource for the embryo (Lopes and Larkins, 1993), the extended expression period from both alleles in the hybrid as compared to one allele in either parent may contribute to the hybrid vigor during seed germination. Interestingly, the heterochronic allelic variation of these genes was more evident when Mo17 and N2 alleles were maternally transmitted and was not obvious when paternally transmitted. These results suggest parental effects on the developmental regulation. One possibility was that there was an interaction of the endosperm with the maternal genotype, which affected the developmental programming. Alternatively, this could be because of the fact that the expression level of some alleles was too low to be detected in some cases.
Another contributing factor to the deviation from allelic dosage expression of BGAF and dz-18 was the allelic variation in the level of expression. Allele expression variation has also been reported in human genes (Yan et al., 2002). This finding suggested the connection of genotype to disease susceptibility, based on changes in gene expression, as opposed to changes in the structure of the encoded protein. Alleles of BGAF and dz-18 exhibited similar expression patterns in hybrids of different genetic backgrounds. It is interesting to note that B73 and S2 were both Iowa Stiff Stalk Synthetic (SSS) lines, and their alleles had higher relative expression in comparison to both Non-Stiff Stalk (NSS) lines, Mo17 and N2. The Stiff Stalk lines are usually selected as the female parent in breeding and the Non-Stiff Stalk lines are generally bred as the male parent. It is possible that the predominantly maternal allele expression of these genes was a consequence of breeding selection in favor of traits for maternal parents, such as yield, kernel size, and stress tolerance, during germination.
Gene-specific imprinting of the nrp1 gene
Genomic imprinting has been implicated to play a role in endosperm development. The imprinted genes in plants can be categorized into two distinct classes according to allele-specific versus gene-specific imprinting. Genes regulated by allele-specific imprinting such as r1 (Kermicle, 1970), 19- and 22-kDa zeins and α-tubulin (Lund et al., 1995a,b) are not essential to seed development. Gene-specific imprinting, however, regulates developmentally important genes (Alleman and Doctor, 2000; Baroux et al., 2002). The known genes that exhibit gene-specific imprinting are MEA, FIS2, FIE in Arabidopsis (Chaudhury et al., 2001; Grossniklaus et al., 1998; Luo et al., 2000) and fie1 in maize (Danilevskaya et al., 2002). Data suggest that all these genes, which are regulated by gene-specific imprinting, may be important in controlling endosperm development.
In this study, we discovered a new imprinted gene, nrp1, which has not previously been described in maize. Our results indicated that nrp1 was regulated by gene-specific imprinting, at least in the genotypes examined. This is the second gene that has been characterized in maize, exhibiting gene-specific imprinting. Furthermore, nrp1 was expressed exclusively in the endosperm, throughout the endosperm development, and peaked at 25 DAP. The endosperm-specific expression of nrp1 is consistent with the notion that imprinted genes are primarily expressed in the endosperm (Alleman and Doctor, 2000; Haig and Westoby, 1989). NAM is a member of a large gene family that are suggested to be transcriptional factors important for plant development (Kikuchi et al., 2000; Souer et al., 1996). The highly conserved domains between maize NRP1, rice NAC, and Petunia NAM proteins suggest that the nrp1 gene may function as a transcription factor. The gene-specific imprinting of nrp1 implicates its important role as a putative transcription factor in endosperm development.
Maternal preference in F1 hybrids gene expression
We found that the number of MLE cDNA fragments was consistently higher than that of PLE cDNA fragments. This was the case for all genotypes through all three developmental stages analyzed, with the exception of hybrid N1/S1 at 10 DAP. As the endosperm has two doses of the maternal genome and one dose of the paternal genome, we examined the possibility of the allelic dosage effect on the estimation of the number of MLE cDNA fragments. If the expression pattern of these MLE genes was confounded with the gene dosage effect, we would expect the average expression level of MLE genes to be higher than that of PLE genes. There was no significant difference in the average expression level of cDNA fragments between MLE and PLE classes in any of the genotypes (data not shown). The results suggested that a gene dosage effect was less likely to be responsible for the preponderance of MLE cDNA fragments.
Data from this and other works suggest that a higher number of MLE cDNA fragments could be as a result of preferential expression of the maternally derived alleles in the hybrid endosperm. Vielle-Calzada et al. (2000) reported the delayed activation of paternal alleles occurring at an early stage of seed development in Arabidopsis. However, the tissues we examined were from much later developmental stages; the maternal preference in gene expression was less likely because of the delay of the activation of paternal alleles. DNA methylation profiles of the maize endosperm showed that maternal alleles were more de-methylated than paternal alleles (M. Lauria, unpublished results), which might indicate that maternal alleles were in a more active state than paternal alleles were.
Phenotypically, kernel progeny from diverse crosses such as maize (maternal) by teosinte (paternal) resembles the maternal parent (maize) in shape and size (Alleman and Doctor, 2000). It appears that genes governing seed size and shape are expressed from maternal alleles (Kermicle, 1978; Schwartz, 1965). Imprinting in plants might represent one level of control of gene expression used in the endosperm to maintain maternal control of kernel growth and development in a tissue that is structurally dependent on a maternal organ (Alleman and Doctor, 2000).
Genome-wide mRNA profiling of maize hybrids and their inbred parents provided insight into the global gene expression of the triploid endosperm in relation to the parental genome dosage. This study demonstrated that the majority of genes expressed in the maize endosperm were regulated in an allelic additive manner. Genomic imprinting and allelic expression variations were responsible for the non-allelic dosage expression of selected genes in the endosperm. The combination of parent-specific and dosage regulations may be the underlying mechanisms of the parental balance requirement of 2 maternal : 1 paternal ratio for normal endosperm development.
Maize (Zea mays L.) inbred lines were either from the public collections or from the collection of Pioneer Hi-Bred International, Inc. (Table 1). These materials are either Iowa SSS lines or NSS lines (Labate et al., 1997). Most of these hybrids are crosses between the two heterotic pools, and represent either current or past commercial hybrids, with the exception of hybrids from the two Stiff Stalk lines N2 and N3. The relatedness of the N2 and N3 by pedigree is approximately 20% and inbred parents of other hybrids are less than 2%. We made the reciprocal crosses among seven inbred lines in the field in 1998 to produce eight hybrids: B73/Mo17 and Mo17/B73, S1/N1 and N1/S1, S2/N2 and N2/S2, N2/N3 and N3/N2. Endosperm tissue was collected at 10, 14, and 21 DAP. We harvested the ears from the field and dissected endosperm tissue from the cob in the lab, froze immediately in liquid N2, and stored at −80°C. We collected all materials for mRNA profiling in 1998. Because of the limited tissue remaining after mRNA profiling, we collected additional tissue samples in 2001 for allele-specific gene expression analyses.
RNA isolation and GeneCalling analysis
We extracted total RNA using TriPure reagent (Roche Molecular Biochemicals, Indianapolis, IN, USA) according to the manufacturer's protocol. We purified Poly(A)+ RNA from total RNA using oligo (dT) magnetic beads (PerSeptive, Cambridge, MA, USA) and quantified by fluorometry. Poly(A)+ RNA was then subjected to GeneCalling analysis as described by Shimkets et al. (1999). GeneCalling analysis consists of the following steps. Double-stranded cDNA is synthesized from the mRNA and digested with 48 different pairs of restriction enzymes (6-bp recognition sites). Adapters are ligated to the cDNA, which is then PCR amplified for 20 cycles using adapter-specific primers. The same 48 pairs of PCR primers were used for all the samples in this study. One of the PCR primers is labeled with FAM. The FAM-labeled PCR products are resolved by high-resolution capillary gel electrophoresis to generate traces showing peaks, whose position and height represent fragment size (in base pairs) and abundance of the cDNA fragment, respectively. The trace data corresponding to each cDNA fragment are used for quantitative comparisons between samples from reciprocal hybrids and inbred parents. For each primer pair, three independent PCRs are made from an individual mRNA sample. One mRNA sample from each genotype was analyzed (three PCR reactions), except for one genotype in which three experimental replicates were profiled (nine PCR reactions). The data were normalized based on the assumption that the majority of the transcripts from a given tissue type remain unchanged among genotypes. As we compared different genotypes within the same developmental stages, we normalized the data across genotypes for each developmental stage. The gene identity of each cDNA fragment is determined by cloning and sequencing the cDNA fragment and is confirmed by a competitive PCR method in which the original PCR reaction is re-amplified in the presence or absence of an excess amount of an unlabeled gene-specific PCR primer.
d/a ratio calculation
In order to obtain a quantitative measurement of the F1 hybrid expression level relative to the average of the parental levels (allelic additive expression level) for a corresponding cDNA fragment, we adapted the d/a ratio from quantitative genetics as a metric. In this measurement, d (dominant gene action) = F1 (hybrid) − Ave (average of the parents); a (additive gene action) = P1 (parent 1) − Ave. In the case of a complete dominant gene action of the P1 allele, F1 = P1, then d/a = 1; d/a = −1 if the other parental allele (P2) is dominant. In the case of additive gene action, F1 = Ave, then d/a = 0. (Using P2 in place of P1 will give the same d/a ratio except that the sign of the ratio is reversed: 1 versus −1). Using this concept, we considered the RNA expression level as a phenotype of each gene and measured the F1 hybrid expression level relative to the additive (allelic dosage) expression. In the endosperm, the maternal parent contributes two doses and the paternal parent contributes one dose to the genetic constitution. Additive allelic expression in the hybrid would give an average expression level (Ave) of 2Pfemale + 1Pmale/3. Therefore, for each cDNA fragment that was different between the parents, we first calculated the deviation of the actual hybrid expression level from the average of the parents, as d = F1 − Ave, and then calculated the deviation of the male parent from the average as a = Pmale − Ave. The d/a ratio was then used to measure the hybrid expression level relative to the average of parental levels. If the hybrid expression level is equal to the average expression level, then d = F1 − Ave = 0, which results in d/a = 0. Therefore, a 0 value of the d/a ratio indicates that the level of expression in the hybrid is the same as the average of the parents, and fits the predicted allelic additive expression. If the hybrid expression deviates from the average expression level and is biased towards the male parent level, then the values d = F1 − Ave and a = Pmale − Ave would be either both negative or both positive, resulting in d/a > 0. Likewise, a value of d/a < 0 will be obtained if the hybrid expression is biased towards the female parent level, where the values d = F1 − Ave and a = Pmale − Ave would be opposite in sign, one negative and the other positive. While the absolute value of the d/a ratio indicates the degree of the deviation from the allelic additive expression, the sign of the d/a ratio indicates the direction of the deviation, maternal or paternal.
Selection of differentially expressed genes between reciprocal hybrids
We compared profiles of inbred parents and reciprocal hybrids: AA/A and BB/B, AA/B and BB/A (Figure 2a). First, we selected the cDNA fragments that were differentially expressed by at least twofold between the parents AA/A and BB/B of individual hybrids. The twofold change as a minimum cut-off was empirically determined, based on CuraGen's statistical analysis of the GeneCalling technology regarding the reproducibility, sensitivity, and false-positive rate, and also a general protocol used in other studies. Those cDNA fragments were further compared in regards to the expression difference between reciprocal hybrids AA/B and BB/A. If the expression level of these cDNA fragments in the hybrid was the same as the maternal parent, we selected these as MLE. With the same principle, we selected paternal-like expressed cDNA fragments.
Allele-specific RT-PCR and WAVE dHPLC analysis
We extracted total RNA using the same protocols as used for GeneCalling. The total RNA was treated with DNase I (Invitrogen, Carlsbad, CA, USA). The first-strand cDNA was synthesized using SuperScriptII (Invitrogen, Carlsbad, CA, USA). We used gene-specific primers to obtain the cDNA from each inbred parent by RT-PCR with Pwo polymerase (Roche, Indianapolis, IN, USA). We then sequenced the PCR products to identify allele sequence polymorphisms between the inbred lines that would allow separation of the two parental alleles on the WAVE dHPLC system (Transgenomic, Omaha, NE, USA). We performed 30-cycle PCR with the cDNA of each hybrid. Thirty-cycle PCR was elected because at lower cycle numbers, both alleles might not be detected.
If a size difference of greater than 1% was present between the two parental alleles for a given gene, we used a size-based separation under non-denaturing (50°C) conditions on the WAVE dHPLC. If the PCR products from the two parental alleles were the same size and contained single nucleotide polymorphisms (SNPs), we ran the samples under partially denaturing (mutation detection) conditions to allow separation of the homoduplexes in the hybrid samples. When using partially denaturing conditions, we heated the PCR reactions to 95°C for 5 min, and allowed to cool slowly to 25°C over a 45-min period in order to allow re-annealing prior to running the samples on the WAVE dHPLC. The optimal temperature for mutation detection must be determined empirically for each gene sequence. Chromatogram traces for each PCR were generated by UV detection. Peak areas corresponding to the homoduplexes and heteroduplexes were calculated by the wavemaker software and used for allelic dosage expression analysis.
Tissue-specific gene-expression analysis of the nrp1 gene
Endosperm and embryo tissues of 16-DAP kernels and ovule tissue from B73 were carefully dissected under a microscope to ensure no cross-contamination of tissue. The root, stalk, leaf, immature ear, and tassel tissues were collected from B73 at the V12 stage. We purified the RNA from each tissue and treated RNA with DNase. We used 1 µg of RNA from each tissue for RT-PCR. Thirty-cycle RT-PCR (protocol as described above) was performed using nrp1 gene-specific primers with cDNA from each tissue type. We performed RT-PCR of the α-tubulin gene using the same cDNA from each tissue as a control of cDNA quality and PCR robustness. We then ran each PCR sample on a gel consisting of 1% Seakem LE agarose (BMA, Rockland, ME, USA), 1× TBE for 90 min at 70 V, and stained with ethidium bromide. Molecular weight marker VIII (Roche, Indianapolis, IN, USA) was used to confirm the size of the amplified genes.
Novel materials described in this publication may be available for non-commercial research purposes upon acceptance and signing of a material-transfer agreement. In some cases, such materials may contain or be derived from materials obtained from a third party. In such cases, distribution of material will be subject to the requisite permission from any third-party owners, licensors, or controllers of all or parts of the material. Obtaining any permission will be the sole responsibility of the requestor. Plant germplasm will not be made available except at the discretion of the owner and then only in accordance with all applicable governmental regulations.
We thank Ben Bowen, Howie Smith, Rudolf Jung, and Wesley Bruce for valuable discussion; Edward Bruggemann, Katie Brown, Andrea Rouse, Marcie Vaughn, Erin McLaughlin, Brian Zeka, Dave Ritland, and Pedro Hermon for various help in conducting the experiments; and Oswald Crasta, John Tobias, and Otto Folkerts at CuraGen Corporation for their support with the GeneCalling. We especially thank James Birchler for valuable discussion and suggestions to the manuscript.